Search

Open-domain event extraction is a fundamental task that aims to extract non-predefined types of events from news clusters. Some researchers have noticed that its performance can be enhanced by improving dependency relationships. Recently, graphical convolutional networks (GCNs) have been widely used to integrate dependency syntactic information into neural networks. However, they usually introduce noise and deteriorate the generalization. To tackle this issue, we propose using Bi-LSTM to obtain semantic representations of BERT intermediate layer features and infuse the dependent syntactic information. Compared to current methods, Bi-LSTM is more robust and has less dependency on word vectors and artificial features. Experiments on public datasets show that our approach is effective for open-domain event extraction tasks.

Spelling error detection serves as a crucial preprocessing in many natural language processing applications. Due to the characteristics of Chinese Language, Chinese spelling error detection is more challenging than error detection in English. Existing methods are mainly under a pipeline framework, which artificially divides error detection process into two steps. Thus, these methods bring error propagation and cannot always work well due to the complexity of the language environment. Besides existing methods only adopt character or word information, and ignore the positive effect of fusing character, word, pinyin1 information together. We propose an LF-LSTM-CRF model, which is an extension of the LSTMCRF with word lattices and character-pinyin-fusion inputs. Our model takes advantage of the end-to-end framework to detect errors as a whole process, and dynamically integrates character, word and pinyin information. Experiments on the SIGHAN data show that our LF-LSTM-CRF outperforms existing methods with similar external resources consistently, and confirm the feasibility of adopting the end-to-end framework and the availability of integrating of character, word and pinyin information., 8 pages,5 figures

In the social network, similar users are assumed to prefer similar items, so searching the similar users of a target user plays an important role for most collaborative filtering methods. Existing collaborative filtering methods use user ratings of items to search for similar users. Nowadays, abundant social information is produced by the Internet, such as user profiles, social relationships, behaviors, interests, and so on. Only using user ratings of items is not sufficient to recommend wanted items and search for similar users. In this paper, we propose a new collaborative filtering method using social information fusion. Our method first uses social information fusion to search for similar users and then updates the user rating of items for recommendation using similar users. Experiments show that our method outperforms the existing methods based on user ratings of items and using social information fusion to search similar users is an available way for collaborative filtering methods of recommender systems.

Microblog is a browser-based platform for web user’s information sharing and communication. With the rapidly increasing of microblog population, its recommendation function becomes necessary. This paper proposes the recommendation by the Latent Dirichlet Allocation topic model, which combines the user interests into the model to meet their needs. We also conduct a comparative analysis between indirect and direct recommendation algorithms. The experimental results show that the indirect recommendation is more effective for the micro-blog recommendation.

In this paper, we offer a geometric framework for the computing of a concept’s conceptual vector based on its similarity position with other concepts in a vector space called concept space, which is a set of concept vectors together with a distance function derived from a similarity model. We show that there exists an isometry to map a concept space to a Euclidean space. So, the concept vector can be mapped to a coordinate in a Euclidean space and vice versa. Therefore, given only the similarity position of a concept, we can locate its coordinate and its concept vector subsequently, using distance geometry methods. We prove that such mapping functions do exist under some conditions. We also discuss how to map non-numerical attributes. At last, we show some preliminary experimental results and thoughts in the implementation of an attribute mining task. This work will benefit attribute retrieval tasks.

Bilingual multiword expression extraction is always a significant problem in extracting meaning from free text. This involves analyzing large amounts of textual information. In this paper we propose a text mining approach to extract bilingual multiword expression. Both statistic and rule-based methods are employed into the system. There are two phases in the extraction process. In the first phase, lots of candidates are extracted from the corpus by statistic methods. The algorithm of multiple sequence alignment is sensitive to the flexible multiword. In the second phase, error-driven rules and patterns are extracted from corpus. For acquired high qualified instances, the manual work with active learning is also performed in sample selection. These trained rules are used to filter the candidates. Bilingual comparisons are used in a parallel corpus. Parts of bilingual syntactic patterns are obtained from the bilingual phrase dictionary. Some related experiments are designed for achieving the best performance because there are lots of parameters in this system. Experimental results showed our approach gains good performance.

In this paper, we present a weakly supervised learning approach for spoken language understanding in domain-specific dialogue systems. We model the task of spoken language understanding as a two-stage classification problem. Firstly, the topic classifier is used to identify the topic of an input utterance. Secondly, with the restriction of the recognized target topic, the slot classifiers are trained to extract the corresponding slot-value pairs. It is mainly data-driven and requires only minimally annotated corpus for training whilst retaining the understanding robustness and deepness for spoken language. More importantly, it allows that weakly supervised strategies are employed for training the two kinds of classifiers, which could significantly reduce the number of labeled sentences. We investigated active learning and naive self-training for the two kinds of classifiers. Also, we propose a practical method for bootstrapping topic-dependent slot classifiers from a small amount of labeled sentences. Experiments have been conducted in the context of the Chinese public transportation information inquiry domain and the English DARPA Communicator domain. The experimental results show the effectiveness of our proposed SLU framework and demonstrate the possibility to reduce human labeling efforts significantly.

For the multiword expression (MWE) extraction, the multiple sequence alignment (MSA) is proposed on the motivation of gene recognition. Because textual sequence is similar to gene sequence in pattern analysis. This MSA technique is combined with error-driven rules, with the improved efficiency beyond the traditional methods. It provides a guarantee for the MWE recall. It uses the dynamic programming method to prevent candidates from combinational explosion, and provides a global solution for pattern extraction instead of sub-pattern redundancy. Consequently, it has accurate measures for flexible patterns. In experiment, some advanced statistical measures are performed for ranking candidates. In the comparison experiment, the MSA approach achieved better results.

In this paper we propose a new word sense disambiguation method called Multi-engine Collaborative Bootstrapping (MCB) that combines different types of corpora and also uses two languages for bootstrapping. MCB uses the bilingual bootstrapping as its core algorithm that leading to incremental knowledge acquisition. The EM model is applied to train parameters in a base learner. The feature translation model is improved by semantic correlation estimation. In addition we use multi-engine selection to produce qualified starting seeds from parallel corpora and monolingual corpora. Those seeds that are generated through unsupervised machine learning approaches can also ensure bootstrapping effectiveness in contrast with manually selected seeds in spite of their different selection mechanisms. Experimental results prove the effectiveness of MCB. Some factors including feature space and starting seed number are concerned involved in our experiments because the EM algorithm is sensitive to starting values. Limitation of resources is also a concern.

For the search engine, error-input query is a common phenomenon. This paper uses web log as the training set for the query error checking. Through then-gram language model that is trained by web log, the queries are analyzed and checked. Some features including query words and their number are introduced into the model. At the same time data smoothing algorithm is used to solve data sparseness problem. It will improve the overall accuracy of then-gram model. The experimental results show that it is effective.

This paper presents a language modeling approach to the sentiment detection problem. It captures the subtle information in text processing to character the semantic orientation of documents as "thumb up" (positive) or "thumb down" (negative). To handle this problem, we propose an idea to estimate both the positive and negative language models from training collections. Tests are done through computing the Kullback-Leibler divergence between the language model estimated from test document and these two trained sentiment models. We assert the polarity of a test document by observing whether its language model is close to the trained "thumb up" model or the "thumb down" model. When compared with an outstanding classifier, i.e., SVMs on movie review corpus, language modeling approach showed its better performance.

In this paper, we present a weakly supervised learning approach for spoken language understanding in domain-specific dialogue systems. We model the task of spoken language understanding as a successive classification problem. The first classifier (topic classifier) is used to identify the topic of an input utterance. With the restriction of the recognized target topic, the second classifier (semantic classifier) is trained to extract the corresponding slot-value pairs. It is mainly data-driven and requires only minimally annotated corpus for training whilst retaining the understanding robustness and deepness for spoken language. Most importantly, it allows the employment of weakly supervised strategies for training the two classifiers. We first apply the training strategy of combining active learning and self-training (Tur et al., 2005) for topic classifier. Also, we propose a practical method for bootstrapping the topic-dependent semantic classifiers from a small amount of labeled sentences. Experiments have been conducted in the context of Chinese public transportation information inquiry domain. The experimental results demonstrate the effectiveness of our proposed SLU framework and show the possibility to reduce human labeling efforts significantly.

This paper proposes a new approach for Multi-word Expression (MWE)extraction on the motivation of gene sequence alignment because textual sequence is similar to gene sequence in pattern analysis. Theory of Longest Common Subsequence (LCS) originates from computer science and has been established as affine gap model in Bioinformatics. We perform this developed LCS technique combined with linguistic criteria in MWE extraction. In comparison with traditional n-gram method, which is the major technique for MWE extraction, LCS approach is applied with great efficiency and performance guarantee. Experimental results show that LCS-based approach achieves better results than n-gram.